本文首先介绍单机百万并发的测试方法和测试结果,然后分析go语言50行代码实现的单机百万并发网络服务器背后的秘密
组网
采用6台2核8G内存的云主机作为client
采用1台4核16G内存的云主机作为server
client端设置
设置系统打开的最大文件数为20万
ulimit -n 200000
修改端口可用范围为1024到65535
echo 1024 65535 > /proc/sys/net/ipv4/ip_local_port_range
单台client虚机建立18万连接
配置单网卡多ip,每个网卡配置三个ip,启动三个client进程,每个client进程指定不同的local ip建立6万连接,总共18万连接
server端配置
设置系统打开的最大文件数为100万
ulimit -n 1000000
设置半连接队列和全连接队列长度
测试过程中出现了一个现象,客户端建立了30000连接,服务端只建立了28570连接
经过排查,原因是:
1 全连接队列满了,如下图,overflowed次数在增加
2 tcp_abort_on_overflow 为0,表示如果三次握手第三步的时候全连接队列满了那么server扔掉client 发过来的ack(在server端认为连接还没建立起来)
tcp_abort_on_overflow为 1,表示第三步的时候如果全连接队列满了,server发送一个reset包给client,表示废掉这个握手过程和这个连接(本来在server端这个连接就还没建立起来)
解决方法:
设置半连接队列长度为10000
echo 10000 >/proc/sys/net/ipv4/tcp_max_syn_backlog
设置全连接队列长度为10000
echo 10000 >/proc/sys/net/core/somaxconn
参考 【转】关于TCP 半连接队列和全连接队列 - sidesky - 博客园
linux内核调优tcp_max_syn_backlog和somaxconn的区别-10931853-51CTO博客
设置conntrack最大连接数
默认net.nf_conntrack_max 为 262144,设置为100万
sysctl -w net.nf_conntrack_max=1000000
tcp最大连接数调优,可参考Linux 内核优化-调大TCP最大连接数 - 简书
最终测试结果
server建立起96万连接
平时ss命令使用最多的是ss -anp,这里需要注意在连接数非常大的时候,指定p参数命令慢的几乎不可用,这里只指定an参数
ss比netstat性能好,参考https://blog.csdn.net/hustsselbj/article/details/47438781
cpu和内存使用情况
cpu大概占用2个核,内存3g
查看cpu硬件信息,cpu的频率为2.4G
查看cpu硬件信息,参考 linux(centos)查看cpu硬件信息命令图解教程 电脑维修技术网
客户端、服务端代码实现
客户端
package main
import (
"flag"
"fmt"
"net"
"os"
"time"
)
var RemoteAddr *string
var ConcurNum *int
var LocalAddr *string
func init() {
RemoteAddr = flag.String("remote-ip", "127.0.0.1", "ip addr of remote server")
ConcurNum = flag.Int("concurrent-num", 100, "concurrent number of client")
LocalAddr = flag.String("local-ip", "0.0.0.0", "ip addr of remote server")
}
func consume() {
laddr := &net.TCPAddr{IP: net.ParseIP(*LocalAddr)}
var dialer net.Dialer
dialer.LocalAddr = laddr
conn, err := dialer.Dial("tcp", *RemoteAddr+":8888")
if err != nil {
fmt.Println("dial failed:", err)
os.Exit(1)
}
defer conn.Close()
buffer := make([]byte, 512)
for {
_, err2 := conn.Read(buffer)
if err2 != nil {
fmt.Println("Read failed:", err2)
return
}
// fmt.Println("count:", n, "msg:", string(buffer))
}
}
func main() {
flag.Parse()
for i := 0; i < *ConcurNum; i++ {
go consume()
}
time.Sleep(3600 * time.Second)
}
服务端
package main
import (
"fmt"
"net"
"os"
"time"
)
var array []byte = make([]byte, 10)
func checkError(err error, info string) (res bool) {
if err != nil {
fmt.Println(info + " " + err.Error())
return false
}
return true
}
func Handler(conn net.Conn) {
for {
_, err := conn.Write(array)
if err != nil {
return
}
time.Sleep(10 * time.Second)
}
}
func main() {
for i := 0; i < 10; i += 1 {
array[i] = 'a'
}
service := ":8888"
tcpAddr, _ := net.ResolveTCPAddr("tcp4", service)
l, _ := net.ListenTCP("tcp", tcpAddr)
for {
conn, err := l.Accept()
if err != nil {
fmt.Printf("accept error, err=%s\n", err.Error())
os.Exit(1)
}
go Handler(conn)
}
}
高性能网络编程的线程模型
TPC
TPC 是 Thread Per Connection 的缩写,指每次有新的连接就新建一个线程去专门处理这个连接请求。
模型特点:
- 采用阻塞式I/O模型获取输入数据
- 每个连接都需要独立的线程完成数据输入,业务处理,数据返回的完整操作
存在的问题:
- 并发数较大时,需要创建大量线程来处理连接,系统资源占用较大
reactor
reactor模式的核心组成包括reactor和线程池。reactor负责监听网络连接的IO是否可读可写,线程池负责具体业务的处理。在高并发的场景下,reactor采用epoll的效率非常高。
模型特点:
- 采用非阻塞I/O,I/O多路复用
- 采用线程池来处理业务
golang GPC模型
GPC 是 Goroutine Per Connection 的缩写,指每次有新的连接就新启动一个golang协程去专门处理这个连接请求。
模型特点:
- 可采用阻塞IO的方式编程
- 每个连接都需要独立的协程完成数据输入,业务处理,数据返回的完整操作
为什么GPC可以支持单机百万并发
GPC模型跟TPC模型看起来非常相似,为什么GPC可以支持单机百万并发呢?
GPC模型、TPC模型比较
- 栈大小:GPC模型中goroutine栈初始大小为4kB,栈的大小可以按需动态增加或减小。而TPC模型中线程默认栈大小为1MB。
- IO模型:GPC和TPC都是阻塞式编程。但是GPC模型底层是非阻塞IO,golang在语言层面将非阻塞IO包装成了阻塞IO(底层实现是非阻塞IO未就绪时,读操作返回EAGAIN,golang运行时系统将协程状态设置为Wait,进行协程的切换)
- 协程、线程的切换: 协程的切换比线程切换要简单的多,可参考linux操作系统笔记(进程)
GPC模型背后的秘密
GPC模型底层实现其实是reactor模型,golang在语言层面将这一模型封装好,可以采用阻塞的方式编码
GPC模型源码分析
golang源码版本为1.9.4
IO线程的源码实现
启动一个线程运行sysmon函数
runtime/proc.go
// The main goroutine.
func main() {
g := getg()
// Racectx of m0->g0 is used only as the parent of the main goroutine.
// It must not be used for anything else.
g.m.g0.racectx = 0
// Max stack size is 1 GB on 64-bit, 250 MB on 32-bit.
// Using decimal instead of binary GB and MB because
// they look nicer in the stack overflow failure message.
if sys.PtrSize == 8 {
maxstacksize = 1000000000
} else {
maxstacksize = 250000000
}
// Allow newproc to start new Ms.
mainStarted = true
systemstack(func() {
//启动线程,运行sysmon函数
newm(sysmon, nil)
})
...........
sysmon的实现
sysmon函数执行netpoll,获得可读写的fd,将fd关联的协程的状态设置为ready
runtime/proc.go
func sysmon() {
// If a heap span goes unused for 5 minutes after a garbage collection,
// we hand it back to the operating system.
scavengelimit := int64(5 * 60 * 1e9)
if debug.scavenge > 0 {
// Scavenge-a-lot for testing.
forcegcperiod = 10 * 1e6
scavengelimit = 20 * 1e6
}
lastscavenge := nanotime()
nscavenge := 0
lasttrace := int64(0)
idle := 0 // how many cycles in succession we had not wokeup somebody
delay := uint32(0)
for {
if idle == 0 { // start with 20us sleep...
delay = 20
} else if idle > 50 { // start doubling the sleep after 1ms...
delay *= 2
}
if delay > 10*1000 { // up to 10ms
delay = 10 * 1000
}
usleep(delay)
。。。。
// poll network if not polled for more than 10ms
lastpoll := int64(atomic.Load64(&sched.lastpoll))
now := nanotime()
if lastpoll != 0 && lastpoll+10*1000*1000 < now {
atomic.Cas64(&sched.lastpoll, uint64(lastpoll), uint64(now))
//netpoll中会执行epollWait,epollWait返回可读写的fd
//netpoll返回可读写的fd关联的协程
gp := netpoll(false) // non-blocking - returns list of goroutines
if gp != nil {
// Need to decrement number of idle locked M's
// (pretending that one more is running) before injectglist.
// Otherwise it can lead to the following situation:
// injectglist grabs all P's but before it starts M's to run the P's,
// another M returns from syscall, finishes running its G,
// observes that there is no work to do and no other running M's
// and reports deadlock.
incidlelocked(-1)
//将可读写fd关联的协程状态设置为ready
injectglist(gp)
incidlelocked(1)
}
}
。。。。。。
}
netpoll的实现
netpoll执行epollWait,获取可读写的fd,返回可读写fd关联的协程
runtime/netpoll_epoll.go
// polls for ready network connections
// returns list of goroutines that become runnable
func netpoll(block bool) *g {
if epfd == -1 {
return nil
}
waitms := int32(-1)
if !block {
waitms = 0
}
var events [128]epollevent
retry:
n := epollwait(epfd, &events[0], int32(len(events)), waitms)
// print("epoll wait\n")
if n < 0 {
if n != -_EINTR {
println("runtime: epollwait on fd", epfd, "failed with", -n)
throw("runtime: netpoll failed")
}
goto retry
}
var gp guintptr
for i := int32(0); i < n; i++ {
ev := &events[i]
if ev.events == 0 {
continue
}
var mode int32
if ev.events&(_EPOLLIN|_EPOLLRDHUP|_EPOLLHUP|_EPOLLERR) != 0 {
mode += 'r'
}
if ev.events&(_EPOLLOUT|_EPOLLHUP|_EPOLLERR) != 0 {
mode += 'w'
}
if mode != 0 {
pd := *(**pollDesc)(unsafe.Pointer(&ev.data))
//将pd关联的协程加入到gp协程链上
netpollready(&gp, pd, mode)
}
}
if block && gp == 0 {
goto retry
}
return gp.ptr()
}
injectglist的实现
injectglist将协程的状态设置为ready状态
runtime/proc.go
// Injects the list of runnable G's into the scheduler.
// Can run concurrently with GC.
func injectglist(glist *g) {
if glist == nil {
return
}
if trace.enabled {
for gp := glist; gp != nil; gp = gp.schedlink.ptr() {
traceGoUnpark(gp, 0)
}
}
lock(&sched.lock)
var n int
for n = 0; glist != nil; n++ {
gp := glist
glist = gp.schedlink.ptr()
//将waiting状态的协程设置为runnable
casgstatus(gp, _Gwaiting, _Grunnable)
globrunqput(gp)
}
unlock(&sched.lock)
for ; n != 0 && sched.npidle != 0; n-- {
startm(nil, false)
}
}
服务端socket实现
net.ListenTCP的实现
ListenTCP调用socket函数,socket函数会通过系统调用创建socket、设置非阻塞、bind、listen
net/sock_posix.go
// socket returns a network file descriptor that is ready for
// asynchronous I/O using the network poller.
func socket(ctx context.Context, net string, family, sotype, proto int, ipv6only bool, laddr, raddr sockaddr) (fd *netFD, err error) {
//sysSocket函数会通过系统调用创建socket,并通过系统调用设置非阻塞
s, err := sysSocket(family, sotype, proto)
if err != nil {
return nil, err
}
if err = setDefaultSockopts(s, family, sotype, ipv6only); err != nil {
poll.CloseFunc(s)
return nil, err
}
//为socket分配文件描述符fd
if fd, err = newFD(s, family, sotype, net); err != nil {
poll.CloseFunc(s)
return nil, err
}
// This function makes a network file descriptor for the
// following applications:
//
// - An endpoint holder that opens a passive stream
// connection, known as a stream listener
//
// - An endpoint holder that opens a destination-unspecific
// datagram connection, known as a datagram listener
//
// - An endpoint holder that opens an active stream or a
// destination-specific datagram connection, known as a
// dialer
// - An endpoint holder that opens the other connection, such
// as talking to the protocol stack inside the kernel
//
// For stream and datagram listeners, they will only require
// named sockets, so we can assume that it's just a request
// from stream or datagram listeners when laddr is not nil but
// raddr is nil. Otherwise we assume it's just for dialers or
// the other connection holders.
if laddr != nil && raddr == nil {
switch sotype {
case syscall.SOCK_STREAM, syscall.SOCK_SEQPACKET:
//listenStream会通过系统调用bind绑定socket地址,通过系统调用listen
//进行socket监听,通过fd.init()函数将fd加入epoll
if err := fd.listenStream(laddr, listenerBacklog); err != nil {
fd.Close()
return nil, err
}
return fd, nil
case syscall.SOCK_DGRAM:
if err := fd.listenDatagram(laddr); err != nil {
fd.Close()
return nil, err
}
return fd, nil
}
}
if err := fd.dial(ctx, laddr, raddr); err != nil {
fd.Close()
return nil, err
}
return fd, nil
Accept的实现
net/fd_unix.go
func (fd *netFD) accept() (netfd *netFD, err error) {
//pfd.Accept会执行accept系统调用,返回新的socket连接,
//并设置新的socket连接为非阻塞
d, rsa, errcall, err := fd.pfd.Accept()
if err != nil {
if errcall != "" {
err = wrapSyscallError(errcall, err)
}
return nil, err
}
//为新的连接分配一个文件描述符
if netfd, err = newFD(d, fd.family, fd.sotype, fd.net); err != nil {
poll.CloseFunc(d)
return nil, err
}
//通过netfd.init(),将accept新返回的socket fd添加到epoll
if err = netfd.init(); err != nil {
fd.Close()
return nil, err
}
lsa, _ := syscall.Getsockname(netfd.pfd.Sysfd)
netfd.setAddr(netfd.addrFunc()(lsa), netfd.addrFunc()(rsa))
return netfd, nil
}
internal/poll/fd_unix.go
// Accept wraps the accept network call.
func (fd *FD) Accept() (int, syscall.Sockaddr, string, error) {
if err := fd.readLock(); err != nil {
return -1, nil, "", err
}
defer fd.readUnlock()
if err := fd.pd.prepareRead(fd.isFile); err != nil {
return -1, nil, "", err
}
for {
//accept函数内部会执行accept系统调用
//将返回的新的socket fd设置为非阻塞
s, rsa, errcall, err := accept(fd.Sysfd)
if err == nil {
return s, rsa, "", err
}
switch err {
//socket全连接队列为空
case syscall.EAGAIN:
if fd.pd.pollable() {
//设置协程状态为wait
if err = fd.pd.waitRead(fd.isFile); err == nil {
continue
}
}
case syscall.ECONNABORTED:
// This means that a socket on the listen
// queue was closed before we Accept()ed it;
// it's a silly error, so try again.
continue
}
return -1, nil, errcall, err
}
}
Read的实现
internal/poll/fd_unix.go
// Read implements io.Reader.
func (fd *FD) Read(p []byte) (int, error) {
if err := fd.readLock(); err != nil {
return 0, err
}
defer fd.readUnlock()
if len(p) == 0 {
// If the caller wanted a zero byte read, return immediately
// without trying (but after acquiring the readLock).
// Otherwise syscall.Read returns 0, nil which looks like
// io.EOF.
// TODO(bradfitz): make it wait for readability? (Issue 15735)
return 0, nil
}
if err := fd.pd.prepareRead(fd.isFile); err != nil {
return 0, err
}
if fd.IsStream && len(p) > maxRW {
p = p[:maxRW]
}
for {
//执行read系统调用
n, err := syscall.Read(fd.Sysfd, p)
if err != nil {
n = 0
if err == syscall.EAGAIN && fd.pd.pollable() {
//socket fd没有数据可读,将协程状态设置为wait
if err = fd.pd.waitRead(fd.isFile); err == nil {
continue
}
}
}
err = fd.eofError(n, err)
return n, err
}
}
Write的实现
internal/poll/fd_unix.go
// Write implements io.Writer.
func (fd *FD) Write(p []byte) (int, error) {
if err := fd.writeLock(); err != nil {
return 0, err
}
defer fd.writeUnlock()
if err := fd.pd.prepareWrite(fd.isFile); err != nil {
return 0, err
}
var nn int
for {
max := len(p)
if fd.IsStream && max-nn > maxRW {
max = nn + maxRW
}
//执行write系统调用
n, err := syscall.Write(fd.Sysfd, p[nn:max])
if n > 0 {
nn += n
}
if nn == len(p) {
return nn, err
}
if err == syscall.EAGAIN && fd.pd.pollable() {
//socket fd不可写,将协程状态设置为wait
if err = fd.pd.waitWrite(fd.isFile); err == nil {
continue
}
}
if err != nil {
return nn, err
}
if n == 0 {
return nn, io.ErrUnexpectedEOF
}
}
}
GPC模型总结
1 新建socket、accept的socket都设置为非阻塞
2.新建socket、accept的socket的fd都加入epoll
- Read、Write采用循环读写,如果返回EAGAIN,将协程状态设置为wait
- io线程定期执行sysmon,通过epollWait获取可读写的fd,将fd关联的协程设置为runablev